Study and analysis of multiresolution based enhancement level set segmentation algorithm for brain mr images

Abstract

The main objective of this thesis is to study and analyse newlinemultiresolution based enhancement level set segmentation algorithm for brain newlineMR images. The accuracy of segmentation and the time for segmentation of newlinetumour cells are constrained by poor contrast and noise in these images. Due to newlinethe poor contrast and presence of noise, like salt and pepper noise, gaussian newlinenoise, etc., MR images show poor visibility. This would affect the diagnosis of newlinediseases by the physicians. Because of this, right treatment plan is also newlineconstrained. So that, contrast enhancement and noise smoothing methods must newlinebe introduced in the level set based segmentation steps. As a result, the newlinesegmentation accuracy would be improved with reduced segmentation time. newlineMethods like filters, Histogram Equalization (HE), Adaptive Histogram newlineEqualization (AHE), etc. can be used for improving the contrast of the images. newlineWavelet Transforms (WT) and Contourlet Transforms (CT) can be used for newlinenoise smoothing of the contrast enhanced images. Once the images are newlineenhanced, they can be segmented. Methods like thresholding, region growing, newlineedge detection, clustering, statistical methods, Active Contour Model (ACM), newlineLevel Set Evolution (LSE) are available for performing segmentation of tumour newlinecells. Among these methods, LSE method could provide better segmentation newlineaccuracy. And, the conventional LSE methods used only Gaussian kernel for newlinesmoothing the images. newline Even though researchers have worked with level set based newlinesegmentation, most of them have not focused on enhancing the images before newlinesegmentation and have not given importance for the segmentation time. newlineThis thesis employs AHE along with WT or CT for improving the contrast of the newlineimages and WT or CT for reducing noise. newline